Support Matrix#
Models#
The following table lists the model packaged in this NIM.
Model Name |
Model ID |
Publisher |
|---|---|---|
Synthetic Video Detector |
synthetic-video-detector |
NVIDIA |
Supported GPUs#
The Synthetic Video Detector NIM performs all video decoding on the GPU using the hardware video decoder (NVDEC).
The NIM is compatible with data center, professional (RTX PRO / RTX), and consumer (GeForce RTX) GPUs based on the NVIDIA Blackwell, Hopper, Ada Lovelace, Ampere, and Turing architectures.
The runtime demuxer provides container support. Hardware decoding support for a codec, profile, pixel format, chroma subsampling, bit depth, and resolution depends on the exact GPU. For per-GPU decode capabilities and per-architecture minimum and maximum resolutions, refer to the NVIDIA Video Encode and Decode GPU Support Matrix and the NVDEC Video Decoder API Programming Guide.
Optimized Configurations#
The following data center and professional (server-class) GPUs have been validated for the Synthetic Video Detector NIM, grouped by architecture and compute capability (CC):
GPU |
Architecture (CC) |
Precision |
|---|---|---|
NVIDIA RTX PRO 6000 Blackwell Server Edition, NVIDIA RTX PRO 4500 Blackwell Server Edition |
Blackwell (12.0) |
FP16 |
B200 |
Blackwell (10.0) |
FP16 |
H100, H200 |
Hopper (9.0) |
FP16 |
L4, L40, L40S |
Ada (8.9) |
FP16 |
A2, A10, A16, A40 |
Ampere (8.6) |
FP16 |
A100 |
Ampere (8.0) |
FP16 |
T4 |
Turing (7.5) |
FP16 |
Other Architectures (Workstation/Consumer GPUs)#
The following table lists additional workstation and consumer GPUs that are supported.
GPU |
Architecture (CC) |
Precision |
|---|---|---|
NVIDIA RTX PRO 6000 Blackwell Workstation Edition, RTX 5090, RTX 5080 |
Blackwell (12.0) |
FP16 |
RTX 6000 Ada Generation, RTX 4090 |
Ada (8.9) |
FP16 |
Note
Other NVDEC-capable GPUs from the Turing architecture onward are expected to work but might not have been explicitly validated for this release.
Software#
This section lists the NVIDIA driver, Docker, and software stack versions required to run Synthetic Video Detector NIM.
NVIDIA Driver and Prerequisites#
NVIDIA driver requirements and other prerequisites for Synthetic Video Detector NIM:
Prerequisite |
Version |
Download and install steps |
|---|---|---|
NVIDIA Graphic Drivers for Linux |
571.21+ |
|
Docker |
latest |
Ubuntu, CentOS, and Debian: https://docs.docker.com/engine/install/; Rocky Linux: https://docs.rockylinux.org/gemstones/containers/docker/ |
NVIDIA Container Toolkit |
latest |
Installation and configuration instructions: https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/latest/install-guide.html |
On bare-metal Linux systems with PCIe topology and multiple GPUs, the NIM creates one model instance per GPU and keeps each request’s shared memory pool and inference on the allotted GPU by default. To use Triton’s dynamic scheduling across GPUs instead, enable IOMMU passthrough (iommu=pt) and set NV_AI4M_SVD_PIN_MODEL_TO_SHM_GPU=false. Refer to Multi-GPU.
Synthetic Video Detector NIM uses the following NVIDIA software platforms:
Components |
Version |
|---|---|
CUDA |
12.8.1 |
cuDNN |
9.8.0.87 |
TRT |
10.9.0.34 |
DeepStream |
8.0 |